diff --git a/doc/src/compute_temp_profile.rst b/doc/src/compute_temp_profile.rst
index b70206781f..e5830f5cf7 100644
--- a/doc/src/compute_temp_profile.rst
+++ b/doc/src/compute_temp_profile.rst
@@ -76,21 +76,28 @@ velocity for each atom.  Note that if there is only one atom in the
 bin, its thermal velocity will thus be 0.0.
 
 After the spatially-averaged velocity field has been subtracted from
-each atom, the temperature is calculated by the formula KE = (dim\*N
-- dim\*Nx\*Ny\*Nz) k T/2, where KE = total kinetic energy of the group of
-atoms (sum of 1/2 m v\^2), dim = 2 or 3 = dimensionality of the
-simulation, N = number of atoms in the group, k = Boltzmann constant,
-and T = temperature.  The dim\*Nx\*Ny\*Nz term are degrees of freedom
-subtracted to adjust for the removal of the center-of-mass velocity in
-each of Nx\*Ny\*Nz bins, as discussed in the :ref:`(Evans) <Evans1>` paper.
+each atom, the temperature is calculated by the formula
+*KE* = (*dim\*N* - *Ns\*Nx\*Ny\*Nz* - *extra* ) *k* *T*/2, where *KE* = total
+kinetic energy of the group of atoms (sum of 1/2 *m* *v*\^2), *dim* = 2
+or 3 = dimensionality of the simulation, *Ns* = 0, 1, 2 or 3 for
+streaming velocity subtracted in 0, 1, 2 or 3 dimensions, *extra* = extra
+degrees-of-freedom, *N* = number of atoms in the group, *k* = Boltzmann
+constant, and *T* = temperature.  The *Ns\*Nx\*Ny\*Nz* term is degrees
+of freedom subtracted to adjust for the removal of the center-of-mass
+velocity in each direction of the *Nx\*Ny\*Nz* bins, as discussed in the
+:ref:`(Evans) <Evans1>` paper.  The extra term defaults to (*dim* - *Ns*)
+and accounts for overall conservation of center-of-mass velocity across
+the group in directions where streaming velocity is *not* subtracted. This
+can be altered using the *extra* option of the
+:doc:`compute_modify <compute_modify>` command.
 
 If the *out* keyword is used with a *tensor* value, which is the
 default, a kinetic energy tensor, stored as a 6-element vector, is
 also calculated by this compute for use in the computation of a
 pressure tensor.  The formula for the components of the tensor is the
-same as the above formula, except that v\^2 is replaced by vx\*vy for
-the xy component, etc.  The 6 components of the vector are ordered xx,
-yy, zz, xy, xz, yz.
+same as the above formula, except that *v*\^2 is replaced by *vx\*vy* for
+the xy component, etc.  The 6 components of the vector are ordered *xx,
+yy, zz, xy, xz, yz.*
 
 If the *out* keyword is used with a *bin* value, the count of atoms
 and computed temperature for each bin are stored for output, as an
@@ -123,10 +130,20 @@ needed, the subtracted degrees-of-freedom can be altered using the
 .. note::
 
    When using the *out* keyword with a value of *bin*, the
-   calculated temperature for each bin does not include the
-   degrees-of-freedom adjustment described in the preceding paragraph,
-   for fixes that constrain molecular motion.  It does include the
-   adjustment due to the *extra* option, which is applied to each bin.
+   calculated temperature for each bin includes the degrees-of-freedom
+   adjustment described in the preceding paragraph for fixes that
+   constrain molecular motion, as well as the adjustment due to
+   the *extra* option (which defaults to *dim* - *Ns* as described above),
+   by fractionally applying them based on the fraction of atoms in each
+   bin. As a result, the bin degrees-of-freedom summed over all bins exactly
+   equals the degrees-of-freedom used in the scalar temperature calculation,
+   :math:`\Sigma N_{DOF_i} = N_{DOF}` and the corresponding relation for temperature
+   is also satisfied :math:`\Sigma N_{DOF_i} T_i = N_{DOF} T`.
+   These relations will breakdown in cases where the adjustment
+   exceeds the actual number of degrees-of-freedom in a bin. This could happen
+   if a bin is empty or in situations where rigid molecules
+   are non-uniformly distributed, in which case the reported
+   temperature within a bin may not be accurate.
 
 See the :doc:`Howto thermostat <Howto_thermostat>` page for a
 discussion of different ways to compute temperature and perform
diff --git a/src/compute.h b/src/compute.h
index 3d6323b3d1..6d0e4bd0f1 100644
--- a/src/compute.h
+++ b/src/compute.h
@@ -108,7 +108,7 @@ class Compute : protected Pointers {
   Compute(class LAMMPS *, int, char **);
   ~Compute() override;
   void modify_params(int, char **);
-  void reset_extra_dof();
+  virtual void reset_extra_dof();
 
   virtual void init() = 0;
   virtual void init_list(int, class NeighList *) {}
diff --git a/src/compute_temp_profile.cpp b/src/compute_temp_profile.cpp
index ff6bc36a08..4b8e77029b 100644
--- a/src/compute_temp_profile.cpp
+++ b/src/compute_temp_profile.cpp
@@ -119,6 +119,9 @@ ComputeTempProfile::ComputeTempProfile(LAMMPS *lmp, int narg, char **arg) :
   nbins = nbinx*nbiny*nbinz;
   if (nbins <= 0) error->all(FLERR,"Illegal compute temp/profile command");
 
+  nstreaming = (xflag==0 ? 0 : 1) + (yflag==0 ? 0 : 1) + (zflag==0 ? 0 : 1);
+  reset_extra_dof();
+
   memory->create(vbin,nbins,ncount,"temp/profile:vbin");
   memory->create(binave,nbins,ncount,"temp/profile:binave");
 
@@ -197,9 +200,10 @@ void ComputeTempProfile::dof_compute()
   natoms_temp = group->count(igroup);
   dof = domain->dimension * natoms_temp;
 
-  // subtract additional d*Nbins DOF, as in Evans and Morriss paper
+  // subtract additional Nbins DOF for each adjusted direction,
+  // as in Evans and Morriss paper
 
-  dof -= extra_dof + fix_dof + domain->dimension*nbins;
+  dof -= extra_dof + fix_dof + nstreaming*nbins;
   if (dof > 0) tfactor = force->mvv2e / (dof * force->boltz);
   else tfactor = 0.0;
 }
@@ -334,14 +338,19 @@ void ComputeTempProfile::compute_array()
 
   MPI_Allreduce(tbin,tbinall,nbins,MPI_DOUBLE,MPI_SUM,world);
 
-  int nper = domain->dimension;
+  double totcount = 0.0;
   for (i = 0; i < nbins; i++) {
     array[i][0] = binave[i][ncount-1];
+    totcount += array[i][0];
+  }
+  double nper = domain->dimension - (extra_dof + fix_dof)/totcount;
+  double dofbin, tfactorbin;
+  for (i = 0; i < nbins; i++) {
     if (array[i][0] > 0.0) {
-      dof = nper*array[i][0] - extra_dof;
-      if (dof > 0) tfactor = force->mvv2e / (dof * force->boltz);
-      else tfactor = 0.0;
-      array[i][1] = tfactor*tbinall[i];
+      dofbin = nper*array[i][0] - nstreaming;
+      if (dofbin > 0) tfactorbin = force->mvv2e / (dofbin * force->boltz);
+      else tfactorbin = 0.0;
+      array[i][1] = tfactorbin*tbinall[i];
     } else array[i][1] = 0.0;
   }
 }
@@ -576,6 +585,12 @@ void ComputeTempProfile::bin_assign()
 
 /* ---------------------------------------------------------------------- */
 
+void ComputeTempProfile::reset_extra_dof() {
+  extra_dof = domain->dimension - nstreaming;
+}
+
+/* ---------------------------------------------------------------------- */
+
 double ComputeTempProfile::memory_usage()
 {
   double bytes = (double)maxatom * sizeof(int);
diff --git a/src/compute_temp_profile.h b/src/compute_temp_profile.h
index f9f2e42156..2d73781e3f 100644
--- a/src/compute_temp_profile.h
+++ b/src/compute_temp_profile.h
@@ -34,6 +34,7 @@ class ComputeTempProfile : public Compute {
   void compute_vector() override;
   void compute_array() override;
 
+  void reset_extra_dof() override;
   void remove_bias(int, double *) override;
   void remove_bias_thr(int, double *, double *) override;
   void remove_bias_all() override;
@@ -47,6 +48,7 @@ class ComputeTempProfile : public Compute {
   int nbinx, nbiny, nbinz, nbins;
   int ivx, ivy, ivz;
   double tfactor;
+  double nstreaming;
 
   int box_change, triclinic;
   int *periodicity;